To compute shape from a sequence of defocused images, we make a connection between the geometrical approach of Favaro et al. and frame theory, interpreting defocusing operators as analysis operators of frames. In order to solve the inverse problem we impose regularization conditions with respect to frame decompositions in a cosparse setting and solve two minimization problems - one for dictionary learning, one for reconstructing the shape.
This presentation is part of Contributed Presentation “CP9 - Contributed session 9”